OHDSI ATLAS 简介

翁春华

哥伦比亚大学医疗信息学系副教授

美国医学信息学院当选成员

Chunhua Weng, PhD, FACMI

Associate Professor of Biomedical

Informatics

Columbia University, New York City

August 21, 2017, Hangzhou, China, OHDSI-China Meeting

讲座结束后，希望您能够

1. 熟悉ATLAS 的适用和功能

2. 使用ATLAS 选择有特定临床特征的病人组并对

他们进行分析

Why OHDSI?

• 观察性研究(Observational Studies)是任何随机临床测试试验

(RCT)的必要先导(Precursor)；前者发现问题，提出假设，积累

证据；后者用最严格的实验设计验证假设。

• 利用OHDSI 的大数据和工具，研究人员可以

o刻画临床特征 (what happened to the patients?)

 糖尿病人吃什么药？那些有并发症？他们对治疗的响应如何？

o基于人群的估计 (what are the causal effect?)

 哪种糖尿病治疗方案最佳？

o基于病人个体的预测 (what will happen to patient X?)

 哪个病人更适合哪种治疗方案？

用观察型研究指导临床测试设计

• 设计应用符合研究的病人条件 (inclusion and exclusion

criteria)

• 进行观察性研究来指导RCT 设计和目标制定

• 展示药物在临床的使用及其安全性

ATLAS 是什么？

• 一个免费网络开源软件工具

• 定义查询病人的条件（比如，近五年

内有二型糖尿病但没有高血压）

• 运用查询条件找到符合条件的病人

• 分析病人的特征

• 对病人特征的描述可分享，重用，可

自动在不同系统间转换 JSON, SQL,

etc.

Atlas 工作流

设计一个研究

（定义目标人

群，比较人群，

结果评估，统

计分析，等等）

创建病人组和

必须的概念集

建立统计分析

流程

输出临床协议

和分析代码

ATLAS 病人定义的抽象模型:

Everything is a concept

有糖尿病

的人群

对应变量和规则：

 有不正常血糖的病人

 有在服用糖尿病相关药物的病人

 有病历里含有二型糖尿病诊断码 (ICD 9, ICD 10, READ, SNOMED) 的病人

 有病历里含有相关文字描述(二型糖尿病，DMII，Type 2 Diabetes)的病人

等等

Insulin, Metformin and

its brand names, etc.

ATLAS 能干什么？

1. 浏览数据源

2. 检索术语 Vocabulary

3. 定义术语集 Concept Set

4. 定义病人组和他们的临床特征

5. 查询数据库找到符合条件的病人组

6. 可视化单独病人的情况

7. 做人群的效果估计

1. 浏览数据源

2. 检索术语 Vocabulary

• ID: a unique concept ID in OHDSI OMOP CDM

• Concept Code: concept identifier in the source

vocabulary

• Class: categories defined by the source vocabulary

• RC: The record count. This will show the number of

records that are coded with this concept in the

• DRC: The descendant record count. The DRC column will

show the sum of all descendant concepts that are coded

in the CDM.

• Domain: categories defined by the OMOP CDM (e.g.,

Condition, Person, Observation, Specimen, etc)

3. 定义术语集 Concept Set

• 什么是 concept set?

They are lists of concepts from the standardized

vocabulary that taken together describe a topic of

interest for a study.

• 为什么要 concept set？

便于随意选择组合概念集以适用于不同场景

• Exclude: Selecting this checkbox will prevent that concept

from being used in the concept set.

• Descendants: Selecting this check box will use the vocabulary

relationships to automatically select all descendants. If this

option is used in conjunction with the exclude option, it will

exclude the current concept and all descendants.

• Mapped: Selecting this check box will use the vocabulary

relationships to automatically select all concepts mapped to

the selected concept.

3. 定义术语集 Concept Set

4. 定义病人组和他们的临床特征

Cohorts

• Concept sets

which are used in

the cohort

definition will

be listed under

the “Concept

Sets” tab

定义病人组

• 主要事件 Primary Event (Start Date)

Cohort definitions can have lots of rules

But the primary event is the bouncer Have to clear this bar

for the rest of the rules to come into play

Besides being the first rule, the primary event is critical

because it sets the

index date

• 符合的条件 Qualifying Criteria

All the other criteria you wish you require of your cohort

members

• 退出条件 Exit Criteria (End Date)

Defines the end date of the individual in the cohort

Index Date

• The patient’s index date (aka cohort start date) is

determined by when they satisfy the primary event

• The cohort start date can be limited to just first time

a patient meets it or you can count every time they meet

• Subsequent criteria are very commonly tied relative to

the index date

Qualifying Criteria

• All the other criteria you wish you require of

your cohort members Noting that it is still the

primary event that will mark their point of entry

in the cohort

• Can have AND or OR logic

• Can apply the same filters as primary event

• Temporal limitations relative to index

Cohorts 其他工具

• Reporting - The reporting tab provides cohort

summarization and visualization tools

• Explore – 可视化病人信息

• Export – 查看源码，XML, JSON, SQL, etc.

可分享的人群组定义

5. 查询数据库找到符合条件的病人组

Profiles, Jobs and Configuration

• Profiles – Patient level information visualization

• Jobs – Jobs those are running in the background

• Configuration - Select the “configuration” menu

item to review the data sources that have been

configured in the source configuration section. This

screen will let you review options. At this time, it

cannot be used to edit the configuration - that must

be done directly in the database.

Export Your Study – Protocol

Export Your Study – R codes

谢谢您的兴趣和加入！